Short answer

July 2026 in one paragraph: Anthropic expanded Claude Cowork to mobile and web, OpenAI answered two days later with ChatGPT Work on the new GPT-5.6 — both agents built to complete multi-step business tasks — and the EU pushed its high-risk AI Act deadline from August 2026 to December 2027. Agents are officially an enterprise product category, and the compliance clock just got longer.

Some months move the industry an inch; July moved it a category. Within 48 hours, the two biggest AI labs shipped competing products with the same thesis — agents that complete business tasks, not chat about them — while Brussels quietly rewrote the compliance calendar that many teams had built their year around.

Below: the three stories that matter if you run a business, each with the original source linked — plus what we'd actually do about them.

Anthropic takes Claude Cowork to mobile and web

Cowork — Anthropic's agent that works directly with files, folders, and apps — left the desktop on July 7: Max subscribers can now start a task at their desk, follow it from their phone, and collect the finished output later. The more interesting signal is the usage data: in a sample of 1.2 million Cowork sessions across 600,000+ organizations, business process work led at 33.4% (reports, reconciling spreadsheets), content and copywriting took 16.4% — and software development was only 8.7%. The agent era is not a developer story; it's an operations story.

techcrunch.com — sourceThe coding agent wars are spilling into the rest of the officeClaude Cowork expands to mobile and web for Max subscribers.

OpenAI ships ChatGPT Work on GPT-5.6

Two days later, OpenAI launched ChatGPT Work: an agent that gathers context across connected apps, breaks a goal into steps, stays on complex projects for hours, and returns finished sheets, slides, docs, and shareable web apps. It runs on the new GPT-5.6 family (Sol, Luna, Terra), and ships alongside a consolidation: the Codex app merges into one ChatGPT desktop app. The two biggest labs now agree on the product thesis — sell finished work, not conversations.

bloomberg.com — sourceOpenAI unveils ChatGPT Work agent to field tasks for hoursThe launch deepens the race for workplace AI tools. forbes.com — sourceOpenAI debuts ChatGPT Work with GPT-5.6Three model variants — Sol, Luna, Terra — power the new agent.

Brussels moves the AI Act goalposts

On June 29, the Council of the EU gave final approval to the Digital Omnibus package: the compliance deadline for stand-alone high-risk AI systems (Annex III — hiring, credit scoring, essential services…) moves from August 2, 2026 to December 2, 2027, with embedded high-risk systems pushed to August 2028. The obligations themselves — risk management, data governance, logging, human oversight — are unchanged, and so are the penalties (up to €15M or 3% of turnover). A reprieve, not a repeal: the companies that use the extra 16 months win.

traverssmith.com — sourceEU agrees to delay key AI Act compliance deadlinesCouncil approval of the Digital Omnibus, June 29, 2026. hklaw.com — sourceUS companies and the EU AI Act's high-risk obligationsWhat Articles 9–17 and 26 require from providers and deployers.

The Arrow read

Agents finish tasks. They still don't know your business.

Cowork and ChatGPT Work make one thing normal: delegating whole tasks to AI. What they can't ship in a box is your context — approved sources, workflows, permissions, and review rules. That layer is built, not bought. And while everyone stares at agents, the quieter shift continues: buyers ask AI who to trust before they ever visit a website. Being the cited answer — GEO — plus one well-scoped agent workflow with human review is still the highest-ROI combination for an SMB in 2026.